一种改进的带有情感信息的词向量学习方法

Improved approach of word vector learning via sentiment information

词语的情感信息对于情感分析任务至关重要,现有大多数基于词向量的无监督学习方法只能对词语的语法语境建模,但忽略了词语的情感信息。针对这一问题,提出了一种结合监督学习和非监督学习的词向量学习方法,既能够获得词语的语义信息又能够获得情感内容。在相关实验中,对词向量分析作了直观的举例对比,并将该方法用于情感分类任务中,通过引入新的评论数据集对该方法进行验证。实验结果表明,融合了语义与情感的词向量方法效果良好,能更为精确地对情感信息进行分类、更为客观地对用户信息进行评价,助力社交网络良性发展。

The sentiment information is crucial to sentiment analysis task. In learning continuous word representation, typical existing unsupervised algorithms only model the syntactic context of words but fail to capture the sentiment information of text. To solve this problem, this paper presented a hybrid method to capture both semantic information and sentiment content in word vector via unsupervised and supervised learning. By applying this method into sentiment classification task, this paper also compares the method from other different methods. Experiments results show that the proposed method can achieve a better result and improve classification performance effectively, which can evaluate clients＇ information kindly and objectively and contribute to other text classification tasks in social network.